Solar Power Forecasting



The transition towards sustainable energy is one of the biggest challanges of mankind. A shift to decentralized renewable energy and a higher energy efficiency is crucial. Solar forecast information is essential for an efficient use, the management of the electricity grid and for solar energy trading.


To provide a short term solar power forecast for the next 48h with a high resolution based on recent weather forecasts from the DWD (Deutscher Wetterdienst).


Trainingdata for the machine leraning model was created by combinbing historical weather data with historical timeseries data from the power output (kW) of a photovoltaic system. Through a machine learning algorithm (feedforward neural network) a model is trained. This model can be used to predict the solar output for the photovoltaic system for the next 48 hours based on new numerical weather forecasts (NWP) of the DWD.


The result of the project was a model that can predict the solar power generation with an RMSE (root mean squared error) of 25%

Project Name: Solar Power Forecasting
Tech Stack: Python, Tensorflow, Keras
Year: 2019
Client:University Project and Client Project